Neural oscillation occurs in many parts of the brain across many species and is often a network phenomenon involving the participation of diverse cell ensembles and brain areas. As such, multi-electrode recordings are seen as key to provide answers regarding mechanisms and functionalities of neural oscillations. Statistically, a natural framework for analyzing oscillatory neural signals is spectral analysis. Although widely used, the application of spectral analysis to multi-channel neural recordings require certain adaptation and expansion of the traditional approach to meet unique challenges posed by the nervous system. First, the brain changes its functional states on rapid time scales (40 to 50 ms, possibly shorter) during cognitive performance. Second, functional couplings among multiple signals are currently assessed by symmetric measures (no directionality) like coherence spectra, but increasingly more elaborate theories of neural oscillation demands that directionality be added to neural interactions. We meet these challenges by (a) developing a MultiVariate AutoRegressive (MVAR) time series modeling approach to spectral analysis which is capable of examining neural signals over analysis windows as brief as 50 to 60 ms and by (b) incorporating Granger Causality spectra into the MVAR approach to evaluate causal influences and directions of driving among multiple neural signals. We propose to analyze two existing datasets from behaving monkeys by the new methodology to explore its applicability and effectiveness. The first dataset consists of local field potentials simultaneously recorded from up to 16 bipolar intracortical electrodes chronically implanted in one hemisphere while the monkey performed a visuomotor GO/NOGO task. The goal here is to study the dynamical organization of a large-scale oscillatory sensorimotor network supporting the maintenance of pressure on a depressed mechanical lever. The second dataset consists of local field potentials recorded from two linear electrodes, each with multiple contacts spanning all the cortical layers, placed simultaneously in V1 and LGN while the monkey performed an intermodal selective attention task (visual versus auditory). The goal here is to study synchronized oscillations between V1 and LGN and the question of top-down influence during visual attention deployment by examining laminar patterns of neural interactions.